[ 
https://issues.apache.org/jira/browse/PHOENIX-4276?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16639861#comment-16639861
 ] 

Josh Elser commented on PHOENIX-4276:
-------------------------------------

{quote}Ideally, I think we should use hbase-metrics-api where it's available 
(1.4+), and hadoop-metrics2 for the older branches.
{quote}
+1
{quote}Hopefully it's not too much work if there's a common interface we can 
use. I think I wrote a pluggable PMetricRegistry in PHOENIX-3807 (this just 
reminded me that we should go back and get that in)
{quote}
Oy, PHOENIX-3807 is giving me lots of regret on things I said I'd do :(. I 
guess i'd pose the question: do we expect to keep doing 1.2/1.3 focused 
releases? I know there are intentions to keep 1.2 releases going, but I think 
1.3 has stalled out already.

> Surface metrics on statistics collection
> ----------------------------------------
>
>                 Key: PHOENIX-4276
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-4276
>             Project: Phoenix
>          Issue Type: Improvement
>            Reporter: Samarth Jain
>            Assignee: Ashish Misra
>            Priority: Major
>
> It would be good to get an insight on how stats collection is doing over 
> time. An initial set of metrics that I can think of would be:
> Time taken to compute stats (reading cells and computing their size)
> Time taken to commit stats per physical table.
> Number of guide posts collected per physical table
> Number of guide posts collected per region.
> Number of regions on which stats collection happened per physical table
> Number of times stats was collected due to major compaction vs update stats 
> per physical table
> If possible, figure out if stats was collected because minor compaction was 
> promoted to major compaction and surface a metric for it.
> Because most of the collection work happens on server side, one option would 
> be to see how HBase's metrics are surfaced (my guess is JMX) and follow the 
> same pattern. Or we could possibly use the hbase-metrics-api module but that 
> is an HBase 1.4 thing. Another option would be see PHOENIX-3807 for some 
> inspiration.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to